Roadside view angle multi-target tracking method

Abstract

The invention relates to a roadside view angle multi-target tracking method, which is characterized in that paired targets are matched, so that the problems of missing detection, false detection and tracking target identity conversion caused by two shielded targets due to too short distance are effectively reduced; furthermore, by adopting a lightweight network and an advanced activation function, a higher calculation speed can be achieved, the precision is kept unchanged while network parameters are greatly reduced, and relatively accurate initialization and observation values can be provided for subsequent target tracking; furthermore, two pairs of detection targets are adopted to calculate an appearance cosine matrix, so that the calculation amount can be reduced, and the target matching speed can be improved; meanwhile, for an object formed by target pairs which are not successfully matched, matching is further carried out through a single target, so that the matching success rate is improved, and the multi-target tracking precision is improved.

Description

technical field [0001] The invention belongs to the technical field of intelligent traffic environment perception, and in particular relates to a roadside viewing angle multi-target tracking method. Background technique [0002] Environmental perception technology is an important part of intelligent transportation, and target tracking technology is a key part of environmental perception. Target tracking technology is divided into single target tracking and multi-target tracking. At present, the single-target tracking technology is relatively mature and has strong universality, but the multi-target tracking technology is not particularly ideal in the face of background occlusion, illumination, and scale changes in complex traffic environments. [0003] Multi-target tracking technology is widely used in various traffic environments, so a multi-target tracking algorithm with good detection technology and data association and matching mechanism is very necessary for complex tra...

AI Technical Summary

Problems solved by technology

Although the number of neural network layers is deepening and data association methods are improving, it is still difficult to overcome the target missed detection, false detection, and tracking targets caused by dense pedestrians, mutual occlusion between vehicles, and small targets in complex traffic environments. The problem of identity conversion; the loss of trajectory due to changes in illumination scale and shooting angle, and the wrong matching problem caused by changes in the appearance state and motion state of the tracking target during motion; excessive calculation due to the deepening of neural network layers Problems such as large and gradient disappearance or explosion

Images